Unlocking Tacit Knowledge with Customer-Grade AI

Most companies that churn from a traditional Customer Success Platform never buy another one. When a category loses most of its customers to nothing at all, the category is the problem. That's from the letter our CEO Keith Frankel published last week on why Totango is now Odie (formerly Totango). It takes a lot to look at a category you've led for fifteen years and say, the model is broken and we're leaving it. Most companies in that position defend the old thing and die a slow death. Instead we chose to rebuild around what post-sales teams actually need now: AI that knows what your best people know. Highly recommend giving the letter a read regardless of whether you're in post-sales or not. Link below.

What if an agent could build its own harness? Most AI agents are static boxes. The intelligence is real — but the harness around it, the shell that decides what the agent can do and how it behaves, only changes when an engineer changes it. Odella's AI employees work differently. They run in secure remote sandboxes where they own their own computer — running programs, writing code, driving a browser, working in real productivity documents. That's the foundation: genuine autonomy, not a chatbot with a to-do list. Then they evolve: → Memory that persists across sessions → Dynamic skills that learn your brand, tone, and processes → The ability to learn from mistakes And the frontier: adapting their own harness. Reshaping their instructions, responding to new events, taking on new tasks — changing how they change, always under human oversight with their safeguards intact. An AI employee that doesn't just do the work — it gets better at being an employee. Read the full thesis 👇 https://lnkd.in/gTN4WU6x

A customer can say "it's fine" and mean five completely different things. Frustration. Resignation. Actual satisfaction. Sarcasm. Or someone just trying to get off the phone. This week reminded us why building tech for customer conversations is hard as it's not really a language problem but a human problem. The real insight is rarely in the words. It's in the pattern behind them. 🔍 One customer saying "it's fine" is a data point. But ten thousand customers saying it the same way, right before they churn? That's a signal. 📊 That's the shift we're chasing at ODIO: from listening to conversations, to listening at scale. Here's the question we keep coming back to: if your customers are quietly telling you something, would you catch it before it becomes a trend? What's a phrase your customers use that means something different every time you hear it? #CustomerExperience #CX #ODIO #ConversationalAI 

78% of organizations are moving into agentic AI, per McKinsey. Almost none have solved the part that actually matters: can you trust what the agents did. A new paper on agentic AI and multi-agent systems names three real gaps: 1/ Objective drift — agents optimizing for the wrong thing without anyone noticing until it's expensive. 2/ Governance lacunae — nobody owns oversight when multiple agents hand work to each other. 3/ No good benchmarks — most evaluation methods weren't built for long, multi-step, real-world tasks. This is exactly why we design AI employees around a role, not a black box — one clear owner, one clear task, human review built into the workflow from day one. Which of these three worries you most in your own AI rollout — drift, oversight, or evaluation?

AI now writes most production code. So what's an AI engineer actually for? Our position: the job is moving up a level — from authoring code to governing the systems that write, run, and improve it. We call it self-learning systems engineering, built on four pillars: transparency, auditability, governance, and infrastructure. The data backs the shift. 67% of orgs now have AI generating or majorly refactoring over half their weekly code (New Relic, 2026). Anthropic's own coding agents went from a 3x research speedup to 52x in under a year. And an MIT study found AI agents wrote 180% more code — but only 30% more shipped to production. Generation is solved. Trust isn't. We wrote up the full research-backed case, including where we think the discipline is headed and how we're building it into Odella. Read it: https://lnkd.in/eh_MTMY2

🚀 New Feature: Assisted Onboarding Your first AI employee sets the tone for everything that follows. Assisted onboarding helps you design the right role, capabilities, and specialisation — built around how your team actually works.

It's a new dawn, it's a new day. Today, Totango becomes Odie: a Customer Knowledge Engine purpose-built for B2B post-sales. Fifteen years and hundreds of enterprise customers into building Customer Success software, we’ve learned the most useful things a company knows about its customers aren't stored in a system. They're in the heads of your best people. That’s the problem with using general-purpose AI for high-stakes customer work. It has your data and none of that context or human perspective. And AI that doesn't know what your best people know shouldn't be anywhere near a revenue-bearing relationship. Customer-grade AI is a context problem, not a data problem. Odie closes the context gap by learning your company’s customer knowledge, and making that knowledge accessible to your entire team, your agents, native Odie add-on apps, or other tools where your team already works. Read more from our CEO Keith Frankel on what’s changed and where we plan to take Odie: https://bit.ly/3ULfqxb

We're opening the doors... Odokai is now available to early adopters and friendly users! After months of building, testing, and refining, we're thrilled to invite you in. Odokai is the AI operating system where you build apps, orchestrate agents, and automate your world through natural conversation. No code required. No limits on imagination. A single AI platform offering a choice of frontier models and open weight models so you stay in control. Individuals & Team plans are hosted managed services while we offer custom Enterprise installations on your cloud or in your datacentre. What early adopters can do today: - Spin up custom apps with AI - Plan and run complex workflows - Connect to your datasources - Create internal AI knoweldgebase experts - Deploy intelligent agents that actually get things done - Connect tools, data, and workflows seamlessly - Use the models in your favourites IDEs - Persist your work in the Odokai workspace - Sync your workspace to Dropbox - Shape the product alongside us **We're looking for curious builders, creative thinkers, and friendly users who want to help define what the next generation of AI-native software looks like.** Your feedback will shape what comes next #AI #EarlyAccess #AIAgents #NoCode #FutureOfWork #Odokai (link in comments)

A new 314-page review of reliable coding agents makes a useful distinction: agents are evaluated as models, but deployed as systems. The model is only one dependency. Reliability also depends on the harness, execution state, retrieval, memory and state management, permissions, review interfaces, resource allocation, verification, and observability. The practical takeaway for teams building or buying AI agents: test the whole path from task construction to safe recovery—not just the model’s benchmark score. If one layer is weak, improvements elsewhere may never show up in the end-to-end outcome. That’s the engineering mindset we bring to AI employees at Odella: connect agents to real tools and processes, make actions reviewable, and design for safe recovery when components fail. Research: https://lnkd.in/e3GP86Bx

Demand for FDEs is increasing exponentially. A recent C&T study projects a 2,100% surge in demand for these specialists by the end of the year. As Chris Taylor, told TechCrunch: “Many FDEs are well equipped to help you roll Claude Code out to your workforce. Very few are capable of building your flagship AI product feature.” That is the gap Ode with Anthropic was built to close, giving enterprises access to the elite engineers who can work alongside their subject matter experts, get rid of operational bottlenecks, and push frontier AI implementations through to production. Read more here: https://lnkd.in/ge4J6bZy